File size: 10,477 Bytes
91126af |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 |
#!/usr/bin/env python3
# Copyright © Niantic, Inc. 2023.
import logging
import shutil
from pathlib import Path
import os
import numpy as np
import argparse
from distutils.util import strtobool
import time
import ace_zero_util as zutil
from ace_util import load_npz_file, compute_knn_mask
from joblib import Parallel, delayed
# from ace_trainer import TrainerACE
# import dataset_io
from PIL import Image
from ace_visualizer import ACEVisualizer
import subprocess
import matplotlib
_logger = logging.getLogger(__name__)
import imageio
def colorize(depth: np.ndarray, mask: np.ndarray = None, normalize: bool = True, cmap: str = 'Spectral') -> np.ndarray:
if mask is None:
depth = np.where(depth > 0, depth, np.nan)
else:
depth = np.where((depth > 0) & mask, depth, np.nan)
disp = 1 / depth
if normalize:
min_disp, max_disp = np.nanquantile(disp, 0.001), np.nanquantile(disp, 0.999)
disp = (disp - min_disp) / (max_disp - min_disp)
colored = np.nan_to_num(matplotlib.colormaps[cmap](1.0 - disp), 0)
colored = (colored.clip(0, 1) * 255).astype(np.uint8)[:, :, :3]
return colored
def _strtobool(x):
return bool(strtobool(x))
def npz2image(npz_file, save_path):
data = np.load(npz_file)
os.makedirs(save_path, exist_ok=True)
image_data = data['images_gt'] if 'images_gt' in data else data['images']
if image_data.shape[1] == 3:
image_data = np.transpose(image_data, (0, 2, 3, 1))
for idx in range(image_data.shape[0]):
img = image_data[idx, :, :]
if img.max() <= 1.0 and img.min() >= 0.0:
img = (img * 255).astype(np.uint8)
elif img.max() <= 1.0 and img.min() < 0.0:
img = (img + 1.0) * 255 / 2.0
img = img.astype(np.uint8)
else:
img = img.astype(np.uint8)
# import pdb; pdb.set_trace()
image = Image.fromarray(img)
image.save(f"{save_path}/gt_image_{idx}.png")
# image.save(os.path.join(save_path, str(idx) + '.png'))
class Options:
def __init__(self, ** kwargs):
self.__dict__.update(kwargs)
if __name__ == '__main__':
# Setup logging levels.
logging.basicConfig(level=logging.INFO)
parser = argparse.ArgumentParser(
description='Run ACE0 for a dataset or a scene.',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--export_point_cloud', type=_strtobool, default=True,
help="Export the ACE0 point cloud after reconstruction, "
"for visualisation or to initialise splats")
parser.add_argument('--dense_point_cloud', type=_strtobool, default=True,
help='when exporting a point cloud, do not filter points based on reprojection error, '
'bad for visualisation but good to initialise splats')
parser.add_argument('--num_data_workers', type=int, default=12,
help='number of data loading workers, set according to the number of available CPU cores')
#Registration parameters
parser.add_argument('--ransac_iterations', type=int, default=32,
help="Number of RANSAC hypothesis when registering mapping frames.")
parser.add_argument('--ransac_threshold', type=float, default=10,
help='RANSAC inlier threshold in pixels')
# Visualisation parameters
parser.add_argument('--render_visualization', type=_strtobool, default=True,
help="Render visualisation frames of the whole reconstruction process.")
parser.add_argument('--render_flipped_portrait', type=_strtobool, default=False,
help="Dataset images are 90deg flipped (like Wayspots).")
parser.add_argument('--render_marker_size', type=float, default=0.03,
help="Size of the camera marker when rendering scenes.")
# parser.add_argument('--iterations_output', type=int, default=500,
# help='how often to print the loss and render a frame')
parser.add_argument('--result_npz', type=str, help='Path to the .npz file containing the raw data (e.g., intrinsic matrices, 3D points, etc.) for visualization')
parser.add_argument('--results_folder', type=Path, default="1217_new", help='path to output folder for result files')
parser.add_argument('--knn_percent', type=float, default=98.0, help='percentage of points to keep in the knn mask')
parser.add_argument('--K_neighbors', type=int, default=10, help='number of neighbors to use for knn mask')
parser.add_argument('--pan_radius_scale', type=float, default=-1, help='factor to control the size of the pan camera')
parser.add_argument('--pan_start_angle', type=int, default=-90, help='start angle of the pan camera')
opt = parser.parse_args()
opt.results_folder = '/'.join(opt.result_npz.split('/')[:-1])
os.makedirs(opt.results_folder, exist_ok=True)
opt.results_folder = Path(opt.results_folder)
image_folder = f'{opt.results_folder}/gt_images'
npz2image(opt.result_npz, image_folder)
opt.rgb_files = f'{image_folder}/*.png'
input_data = load_npz_file(opt.result_npz)
if 'pts_mask' not in input_data.keys():
start_time = time.time()
input_data['pts_mask'] = compute_knn_mask(input_data['pts3d'], opt.K_neighbors, opt.knn_percent)
end_time = time.time()
_logger.info(f"Computing knn mask took {end_time - start_time:.2f} seconds.")
npz_with_mask = f"{opt.results_folder}/points_with_mask.npz"
np.savez_compressed(npz_with_mask, **input_data)
opt.result_npz = npz_with_mask
best_seed = 4
iteration_id = zutil.get_seed_id(best_seed)
# depth = input_data['pts3d'][0,...,-1]
poses = input_data['cam_poses']
H, W = input_data['pts3d'].shape[1:3]
for idx, (pts3d, pose) in enumerate(zip(input_data['pts3d'], poses)):
w2c = np.linalg.inv(pose)
R = w2c[:3,:3]
T = w2c[:3,-1]
pts3d_c2w = np.einsum('kl, Nl -> Nk', R, pts3d.reshape(-1,3)) + T[None]
disp = pts3d_c2w[...,-1]
disp = disp.reshape(H, W)
disp_vis = colorize(disp)
os.makedirs(f"{opt.results_folder}/depth", exist_ok=True)
imageio.imwrite(f"{opt.results_folder}/depth/depth_vis_{idx}.png", disp_vis)
print(f"Visualized depth map is saved to {opt.results_folder}/depth/depth_vis_{idx}.png.")
register_opt = zutil.get_register_opt(rgb_files=opt.rgb_files, result_npz=opt.result_npz)
modify_options = {
'rgb_files': opt.rgb_files,
'render_visualization': opt.render_visualization,
'render_target_path': zutil.get_render_path(opt.results_folder),
'render_marker_size': opt.render_marker_size,
'render_flipped_portrait': opt.render_flipped_portrait,
'session': f"{iteration_id}",
'hypotheses': opt.ransac_iterations,
'threshold': opt.ransac_threshold,
'hypotheses_max_tries': 16,
'result_npz': opt.result_npz,
'only_frustum': True,
'pan_radius_scale': opt.pan_radius_scale,
'pan_start_angle': opt.pan_start_angle
}
for k, v in modify_options.items():
setattr(register_opt, k, v)
reg_state_dict_1 = zutil.regitser_visulization(register_opt)
scheduled_to_stop_early = False
prev_iteration_id = iteration_id
register_opt_iter = zutil.get_register_opt(rgb_files=opt.rgb_files, result_npz=opt.result_npz)
modify_options = {
'rgb_files': opt.rgb_files,
'render_visualization': opt.render_visualization,
'render_target_path': zutil.get_render_path(opt.results_folder),
'render_marker_size': opt.render_marker_size,
'render_flipped_portrait': opt.render_flipped_portrait,
'hypotheses': opt.ransac_iterations,
'threshold': opt.ransac_threshold,
'hypotheses_max_tries': 16,
'result_npz': opt.result_npz,
'state_dict': reg_state_dict_1,
'only_frustum': False,
'pan_radius_scale': opt.pan_radius_scale,
'pan_start_angle': opt.pan_start_angle
}
for k, v in modify_options.items():
setattr(register_opt_iter, k, v)
reg_state_dict_2 = zutil.regitser_visulization(register_opt_iter)
_logger.info("Rendering final sweep.")
final_sweep_visualizer = ACEVisualizer(
zutil.get_render_path(opt.results_folder),
flipped_portait=False,
map_depth_filter=100,
marker_size=opt.render_marker_size,
result_npz=opt.result_npz,
pan_start_angle=opt.pan_start_angle,
pan_radius_scale=opt.pan_radius_scale
)
poses = [final_sweep_visualizer.pose_align(pose) for pose in final_sweep_visualizer.cam_pose]
rgb_path = f'{opt.results_folder}/gt_images'
rgb_files = os.listdir(rgb_path)
pose_dict = {str(rgb_file): 0 for rgb_file in rgb_files}
pose_iterations = [0 for _ in range(len(rgb_files))]
# import pdb; pdb.set_trace()
final_sweep_visualizer.render_final_sweep(
frame_count=150,
camera_z_offset=4,
poses=poses,
pose_iterations=pose_iterations,
total_poses=len(pose_dict),
state_dict=reg_state_dict_2,)
_logger.info("Converting to video.")
# get ffmpeg path
ffmpeg_path = shutil.which("ffmpeg")
# run ffmpeg to convert the rendered images to a video
ffmpeg_save_cmd = [ffmpeg_path,
"-y",
"-framerate", "30",
"-pattern_type", "glob",
"-i", f"{zutil.get_render_path(opt.results_folder)}/*.png",
"-c:v", "libx264",
"-pix_fmt", "yuv420p",
str(opt.results_folder / "reconstruction.mp4")
]
subprocess.run(ffmpeg_save_cmd, check=True)
print(f'The render result video is saved to {opt.results_folder}/reconstruction.mp4.')
if opt.export_point_cloud:
# import pdb; pdb.set_trace()
import trimesh
pts = final_sweep_visualizer.pts3d.reshape(-1, 3)
image = final_sweep_visualizer.image_gt.transpose(0, 2, 3, 1).reshape(-1, 3)
clr = (( image + 1.0 ) / 2.0 * 255.0).astype('float64')
cloud = trimesh.PointCloud(pts, colors=clr)
cloud.export(f"{opt.results_folder}/point_cloud.ply")
print(f'The point cloud is saved to {opt.results_folder}/point_cloud.ply.')
|